A Scalable Multi-level Distributed System-Level Diagnosis

Abstract

The purpose of distributed system-level diagnosis is to have each fault-free nodes determine the state of all nodes of system. The paper presents a Multi-level distributed system-level diagnosis, which considers the problem of achieving scalability and performance tuning for distributed diagnosis. Existing work is aimed to reduce either diagnosis latency or network utilization but scales poorly. A diagnosis algorithm, called Multi-level DSD, is presented to provide scalability, which controls both latency and network utilization in fully connected networks. The algorithm is scalable in the sense that it is possible to diagnose system with large number of processing elements (nodes) by tuning diagnosis parameters. The diagnosis algorithm allows tuning of diagnosis performance to lever latency message cost trade-off. Multi-level DSD divides system in clusters of nodes, where each cluster is either a single node or a group of clusters. Cluster diagnoses itself by running a cluster diagnosis algorithm between its sub clusters. Clusters at each level runs same cluster diagnosis algorithm.